MIT Department of Electrical Engineering & Computer Science

SPECIAL SEMINAR

Monday, March 13, 1995
Marlar Lounge, Room 37-252

Refreshments at 1:45 PM
Talk at 2:00 PM

Digital Sequence Estimation in the Presence of Parametric Uncertainty

Keith M. Chugg
University of Southern California

It is well known that, when all system parameters are known, Maximum Likelihood Sequence Estimation can be implemented via the Viterbi Algorithm (VA) to provide optimal digital data detection for applications such as intersymbol interference (ISI) channels, trellis- coded modulation, and/or multiuser detection. However, if the channel characteristics are unknown or time-varying, the VA cannot be implemented directly. This talk will begin with a brief summary and classification of the wide variety of algorithms which have recently been introduced to perform joint parameter and data estimation, including Per-Survivor Processing (PSP), which associates a parameter estimate with each survivor in the trellis.

The main focus of this talk will be on the problem of performing joint maximum likelihood sequence and ISI channel estimation and providing the theoretical foundation for PSP and related algorithms. The required front-end processing to obtain sufficient statistics will be described, with practical versions viewed as generalizations of Forney's Whitened- Matched Filter (i.e., the known-channel front-end processing). Recursive computation of the likelihood metric allows the problem to be viewed as tree-search which is not solved by the VA. This formulation provides the theoretical foundation for the general technique of PSP, and suggests natural extensions. Performance in both the tracking and acquisition modes will be discussed.